Bạn có chắc chắn muốn xóa bài viết này không ?
Fuzzy Logic and Machine Learning
Fuzzy Logic and Machine Learning
First of all: I apologize everyone for my writing in English. I come to this site because someone of Daynhauhoc.com recomended me this forum if I want to write advanced themes. My name is Joe Nartca, Stanford MS graduate, 42 yrs old American who speaks and understands Vietnamese quite good (I believe).
I am sure that everyone of us is aware of the avalanche of new technologies like Machine Learning, Data Mining, Big Data, Artificial Learning, etc. and as Gorbachev said "Life punishes those who delay." Every nation on this planet Earth races for AI, ML, BD, ... China challenges the US and makes some greatest progress in Computer Technology and AI as well (e.g. China's fastest supercomputer). Machine Learning (ML) is discussed everywhere and probably everyone has tried to do some research on this topic.
What is ML really? Let me quote this definition of Expert System
Machine learning is an application of artificial intelligence (AI) that provides systems the ability to automatically learn and improve from experience without being explicitly programmed. Machine learning focuses on the development of computer programs that can access data and use it learn for themselves.
The process of learning begins with observations or data, such as examples, direct experience, or instruction, in order to look for patterns in data and make better decisions in the future based on the examples that we provide. The primary aim is to allow the computers learn automatically without human intervention or assistance and adjust actions accordingly.
The term "Machine learning focuses on the development of computer programs that can access data and use it learn for themselves." sounds harmlessly, but very meaningfully. ACCESS DATA? Is not that what we daily do with SQL-MySQL? Yes and no. Yes because our daily data come from a database and they are accessible by SQL formulations. No because the data we deal with are not the normalized data, but real raw data which are chaotic and unspecific. For example: Image, sound, flavor and odor. They are the data we absorb daily and these data are omnipresent around us, in our nearest environment. Statistically we absorb 75% our data around us (viewing, hearing, talking, smelling and tasting) and the rest comes from the web or other media (papers, radio or telly). From this viewpoint it's hopelessly impossible to keep these raw data in any database and to evaluate them spontaneouly with SQL or any other conventional tool.
The solution can be, as usual, found in the past. Fuzzy Logic (FL) was introduced by Lotfi Zadeh in 1965 (HERE). This funny image says it all:
It's about the Significance. The Essence of data. Instead of a verbose explanation or a long decription a brief note or a short expression that signifies or conveys enough the content. Even the note or expression is vague or fuzzy the receiver is in a favorable position to combine or to interprete it into a wholeness.
The word fuzzy is mentioned. Also Fuzzy Logic or FL for short is the key solution for AI-driven Machine Learning. Supervised, Unsupervised, Semi-Supervised or Reinforcement machine learning.
The data are abundant and steadily changing. It's hard to cope with the changes and the new incomings. For example, Facebook or Google has to cope with millions of users who come and go and click. The data they generated are vast. So tremendous that any conventional tool has to capitulate, to give up. To cope with such tremendous data waves, or in a buzzword-terminology Big Data, Google "Data-mines" the data. How? Its tool is the well-known algorithm: Map Reduce Algorithm (MRA). However, MRA cannot solve every information. MRA can reduces "Big Data" to a small quantity, but not the meaning of the data. And that only Fuzzy Logic can do. Example:
if weather is hot then the sky is sunny
if the weather is fair then the sky is cloudy
if the weather is rainy then the sky is wet
All these expressions are vague, unpricise, but the message (i.e the essence) is clear: Hot, fair and rainy are vague. Sunny, cloudy and wet are vague, too. But their significance is perceivable.
FL bases on vagueness and vagueness can be mathematically modelled on Probability (e.g. Student Distribution) as a Pyramid or a Trapezoid
The data (or events) are quantified to a sample which could be valid or invalid among the partaken regions (or sections). The quantification is called Fuzzification and the reverse process is Defuzzification. For example the taken sample is 27° Celsius (Hot Pyramid, between 25 and 30), the fuzzified value is then ~0.4 (the cut of a vertical line at 27° and the left line of the Hot-Pyramid). The expression:
if weather is hot then the Sky is sunny
The left expression is true, the fuzzy trueness of hot is 0.4, also 0.4 point is added to sunny on the right side. Fuzzy conditioning does not have an alternative (no else). Hence, an alternative must be expressed by another conditioning (if weather is fair then ...) The sum of all points are computed into areas (left&right: Bisection) or barycenter (Centroid). The Calculation is called Defuzzification and calculated result is the defuzzified final outcome.
MachineLearning with DataMining of BigData using FuzzyLogic is an advanced brandnew field...and it still has a long way to go,
PS. If anyone wants to play with FuzzyLogic (s)he can download the package written in JAVA from HERE.